June 8, 2026

AI for CPA firms: use cases across the audit lifecycle

Amanda Waldmann

Increasing trust with AI for audit and advisory firms.

Key Insights:

Most AI in audit today is task-level help: copilots and chat tools that speed up individual steps but leave the engagement model untouched. Capacity gains are real but capped.
The bigger shift is moving repeatable execution off the practitioner's desk entirely. The hours that used to go to evidence chasing and first-draft testing go to review and judgment instead.
That shift plays out across the full lifecycle: planning, evidence intake, testing, review, and reporting. The lifecycle, not the task, is what the operating model changes.

A senior is renaming PDFs in a shared drive. Two staff are reconciling evidence versions across email threads. The planning memo still hasn't cleared review, substantive testing was supposed to start Monday, and half the budget is already gone before the work that actually requires professional judgment has started.

That's the gap AI closes, and where the next wave of audit work happens. This article covers where AI shows up across the audit lifecycle today, how the engagement model shifts when agents take on the execution and practitioners take on the review, and where Field Agents fit into that shift.

Why AI matters for CPA firms now

The profession is working through a sustained talent shortage, and the pipeline of new graduates has not kept pace. The BLS projects about 124,200 annual openings for accountants and auditors each year through 2034. Demand is growing. Supply is not. At the same time, deficiency rates in PCAOB inspection cycles remain elevated. The math on hiring more people doesn't work, and the quality bar isn't moving.

AI helps where the work is most repetitive: drafting first passes of planning memos, sorting through client evidence, running test procedures across populations, reconciling data into reports. Those are the tasks that absorb senior hours without requiring senior judgment. When agents take that work off the practitioner's desk, the practitioner gets back the time the engagement actually needed from them, with review and conclusion still firmly in their hands.

Two categories of AI in audit today

Most AI in audit today is assistive: it sits next to the practitioner and speeds up the steps they're already doing. The real capacity unlock comes from agentic AI, where the work moves off the practitioner's desk and they step in to review. Fieldguide is built for both, packaged as AI Assist and Agent Workforce.

AI Assist sits next to the practitioner inside the engagement. They open a workpaper, ask a question against a document, get column-level outputs across a sheet, draft a memo faster. The work still happens at the practitioner's pace, but each step takes less time. Useful, but the engagement still gets run the same way.
Agent Workforce changes who does the first pass. The practitioner sets direction and Field Agents execute the repeatable work across the engagement: reviewing client evidence, drafting workpapers, running test procedures, surfacing exceptions. The practitioner reviews the output, applies judgment, and signs off. The engagement runs on a different model: agent-led execution, human-led judgment.

Here's how each shows up across the audit lifecycle, from planning through reporting.

Planning and risk assessment

Scoping, walkthroughs, control design documentation, and risk assessment take days of senior staff time on every engagement, much of it spent rebuilding artifacts that look remarkably similar to last year's.

A senior can draft individual sections faster with AI in the loop: a memo here, a walkthrough narrative there, a question set for a control owner. The bigger change comes when agents pick up the planning workpapers themselves, drafting risk profiles and walkthrough documentation grounded in firm methodology and prior-year work. The output is a first draft aligned to how your firm actually works.

The engagement team still sets scope and materiality, and still reviews the planning memo. What changes is the starting point. The hours between "engagement accepted" and "planning memo reviewed" can compress from days to a fraction of that, and risk assessment follows the same arc: the engagement team reviews a risk profile built from the engagement context instead of building one from scratch.

Client requests and evidence intake

Evidence collection drains engagement budgets quietly. Renaming PDFs, chasing missing files, reconciling versions across email threads and spreadsheets: by the time substantive testing starts, half the budget is gone.

On the audit side, agents handle the work of request drafting, evidence analysis, evidence linking, and extraction: drafting precise PBC requests, reviewing what comes back, and pairing each file with the procedure it supports. On the client side, the back-and-forth runs through Fieldguide's Client Hub: clients see what's outstanding, submit documents, and track requests in one place, without the email and spreadsheet traffic that bogs the process down.

Timing changes too. Traditionally, evidence arrives in a batch and sits untouched until fieldwork starts weeks later. In the new model, agents review documents the moment a client uploads them, flagging missing items and inconsistencies at intake rather than during fieldwork weeks later. Substantive testing stops starting from behind.

Substantive testing and controls testing

Testing is where the agent model has its clearest advantage, because testing is where the most repeatable execution lives.

Practitioners can move through individual tests faster with AI in the loop: column-level outputs, document Q&A, focused search. The bigger change comes when agents run the test workflow itself. Inside Fieldguide, that work is organized around the kinds of audit engagements firms actually run:

For Profit
Not For Profit
Employee Benefit Plans
Lending and Regulatory
Investment Institutions
Request and Evidence

Agents review client-submitted evidence, execute test procedures across the population, flag exceptions, and document draft results. They pull data from supporting documents and link it back to the workpapers it supports, at the scale of the engagement rather than one document at a time. The repeatable extraction and matching work comes off the senior's desk, and practitioners spend their hours on the exceptions.

Judgment stays with the practitioner. Risk-based sampling remains auditor-directed: the auditor determines the methodology, and the agents handle the population analysis and procedure execution within it. The same pattern holds in controls testing, from access rights verification to segregation-of-duties review. Agents execute the defined procedures, draft the documentation and exception flags, and the practitioner reviews and concludes.

Review, documentation, and workpaper management

Review consumes the most expensive hours on the engagement, and it's where deficiencies turn into inspection findings. The PCAOB has taken enforcement action on engagement quality review failures, and most of those failures trace back to issues the engagement team should have caught before the partner saw the file.

Agents change where in the workflow those issues surface. As work moves through the engagement, agents flag exceptions, judgment calls, and elevated-risk areas directly in the workpapers: missing evidence, inconsistent values, control gaps, populations that don't tie. By the time a manager or partner opens the file, the agents have already marked the issues, with a clear pointer to the supporting evidence. Review time shifts from hunting for problems to evaluating the ones the agents have raised.

Catching issues early is half the equation. The other half is giving the team space to validate agent output before it moves up the chain. A dedicated workspace lets the preparer review and append to that output before passing it to the manager, and a kanban-style board shows status across the engagement at a glance. The preparer's work still happens; it happens earlier, on better drafts, with a clear record of what the agent did and what the preparer confirmed.

Documentation holds up to review because every agent run produces an audit trail of inputs, outputs, and reasoning, with citations back to source documents on the surfaces that support them. The workpaper file carries a record of how outputs were generated and what the preparer confirmed.

Reporting and deliverables

Reporting is where manual reconciliation quietly burns hours that never show up in realization analysis.

At the section level, AI speeds up the writing work: drafting language, summarizing findings, formatting tables. The bigger change is in the data flow itself. For financial audit specifically, agents tie financial statement preparation directly to the trial balance, which cuts the manual reconciliation between workpapers and final deliverables that quietly burns the most hours. Across the rest of audit, agents pull workpaper data into report templates and the engagement team reviews and approves the final report.

The operational gain is a shorter reporting workflow. GenAI drafting and standardized report generation are starting to show up across audit workflows, and in the agent model, the data connection is built into the workflow rather than bolted on at the end.

Quality and judgment in this workflow

Quality is the through-line across every phase above. Agents and practitioners build quality into the engagement as the work happens, instead of relying on review hours at the end to catch what got missed.

Professional skepticism shifts with it. Instead of catching missed items buried in workpapers, the team spends more time evaluating the exceptions the agents have already flagged. The inputs to judgment are better, and the time spent on judgment goes further.

Oversight matters as much as function. Fieldguide is built around AI governance and control effectiveness, including SOC 2 Type 2 attestation, ISO 42001 AI governance certification (among the first audit and advisory platforms to achieve it), and AIUC-1 certification. For audit partners who carry personal responsibility for firm practices, platform governance credentials are part of that professional responsibility.

The operating model that runs on this

Fieldguide is an end-to-end AI-native platform purpose-built for audit and advisory, with AI Assist and Agent Workforce covering the audit lifecycle on a single system. Field Agents execute the repeatable work, practitioners review and apply judgment, and firm methodology runs through both. That's the difference between layering AI onto the workflows firms already have and changing how engagements actually run. Teams can see the model in practice by booking a demo with Fieldguide. AI features change how work moves through an engagement; final approval stays with the engagement team.